Table 2.

Automated de-identification system performance by identifier type in the average-density identifier corpus consisting of 50 family practice progress notes (22 525 words)

PHI type	N PHI instances in corpus	N PHI instances replaced by system	System PHI recall*	N residual PHI instance in corpus	Reasonable opportunity to test HIPS?†
A	B	C	D	E	F
HIPAA PHI
Pat. name	59	27	46%	32	Yes
Age	50	50	100%	0	No
Phone #	3	3	100%	0	No
Address	3	0	0%	3	No
Date	228	194	85%	34	Yes
MRN	0	0	NA	0	No
Acct. #	0	0	NA	0	No
Other ID #s	0	0	NA	0	No
ALL HIPAA	343	274	80%	69
OTHER PHI
MD name	53	4	8%	49	No
Org. name	63	2	3%	61	No
ALL OTHER	116	6	5%	110

^*A suboptimal training set was used to degrade system recall, thereby increasing residual PHI for experimental purposes.

^†Criteria for inclusion in the detection experiment were system recall (col. D) ≥ ∼0.5 and N residual instances (col. E) ≥ ∼10.